There's a conversation happening in boardrooms right now that I find deeply misguided. It goes something like: "We need an AI strategy." What they usually mean is: we need to be seen doing something with AI before our competitors do.
That's not a strategy. That's anxiety dressed up as planning.
The operators I've seen use AI well — and I've been watching this closely across my portfolio — share one thing in common: they started with the problem, not the technology. They asked "where are we slowest, most error-prone, most reliant on manual judgement?" and then looked at whether AI could address that specifically.
At Revolut, speed of execution was always a competitive weapon. If AI had been as capable then as it is today, the places I'd have deployed it first wouldn't be customer-facing — they'd be internal. Compliance review. Risk modelling. Localisation. The unglamorous operational work that compounds over time.
The operators who will win the AI era aren't the ones who move fastest. They're the ones who make the most durable choices — who embed AI into their core operations in ways that are hard to unwind and impossible to replicate quickly. Infrastructure beats features, every time.
If you're thinking about where to start: find the workflow in your company that is highest-frequency, most rule-bound, and most expensive when it goes wrong. That's your first deployment. Everything else follows.